"How much does a custom AI chatbot cost?" is a fair question with an unhelpful default answer ("it depends"). Here's a more useful answer: a real cost range, broken down by phase, plus what actually moves the number.
Numbers below reflect the general market for custom, integrated chatbot projects — not template chatbot builders. Your exact quote should come from a discovery call, not a website.
The rough range
Small, focused MVP (one channel, one knowledge source, minimal integration): $8k–$15k, 4–6 weeks.
Medium production build (multi-channel, RAG over multiple sources, CRM + helpdesk integrations, evaluation): $15k–$40k, 8–12 weeks.
Larger enterprise build (agent workflows, SSO, complex permissions, multi-region deploy): $40k+, 12+ weeks.
Ongoing retainer (tuning, features, monitoring): from $3k/month.
Where the money goes
Discovery + architecture: 5–10%. Understanding the use case, scoping data, agreeing success criteria. Skipping this is where most bad chatbot projects go wrong.
Data preparation: 15–25%. Ingesting, cleaning, chunking, and evaluating retrieval quality. The unglamorous work that determines answer quality.
Development + integration: 40–60%. Building the chatbot, wiring in CRM/helpdesk/channels, prompt design, guardrails.
Evaluation + iteration: 15–20%. Building the eval suite, running against real questions, fixing what's broken.
Deploy + monitoring: 5–10%. IaC, observability, cost dashboards, runbooks.
What increases the cost
More knowledge sources with different formats (unstructured PDFs, scanned docs, multiple wikis).
More integrations, especially with legacy systems that don't have good APIs.
Permission complexity (per-document ACLs, SSO integration, tenant isolation).
Multi-language support with high-quality answers in each language.
Regulated industries (healthcare, finance) with compliance requirements.
Ongoing running costs
LLM inference: varies by traffic, model tier, and how much context you send. Typical business chatbot: $50–$500/month at moderate traffic; enterprise usage can be higher.
Vector database: often free at small scale (pgvector on your existing Postgres); $50–$500/month for managed services at scale.
Hosting + observability: modest — $50–$300/month typically.
Retainer for tuning: from $3k/month if you want us to keep improving it; $0 if your team takes it over.
How to keep the cost down
Start narrow. One channel, one clearly defined use case. Prove value, then expand.
Use existing docs — don't invent a new knowledge base for the chatbot.
Pick integrations that actually change customer outcomes; skip the vanity ones.
Design for cost from day one: cheap model for easy questions, premium only for hard ones.
Frequently asked questions
Conclusion
Custom AI chatbots aren't cheap — but a good one earns its cost back fast on deflected tickets, captured leads, and freed staff time. If a vendor quotes without a discovery call, be suspicious.